The ABINIT Users Mailing List ( CLOSED )

[Anglade Pierre-Matthieu <anglade@gmail.com>]

Hi,

This won't answer your question but please consider the following:
Molecular
dynamics is chaotic and you'll never ever get exact results. Whatever
precision you have for your computation you will always get results
very differents (after even a not so big number of steps) from what a
non arbitrary precision computation would gives. 
Then I would say that you can behaves in two different ways: either you
choose to consider that molecular dynamics is not a suitable technic
for your problem; or you decide not to mind that much about the
differences resulting from a  small  error on your atoms
phase space coordinates at the first restart step.

regards

PMA

On 9/23/05, ngalamba@cii.fc.ul.pt <ngalamba@cii.fc.ul.pt> wrote:

Dear ABINIT users

I am trying to use ABINIT to carry molecular dynamics simulations (N,V,E) (IONMVOV 6) of simple ionic melts ( e.g. NaCl).
Because it is hard to get a complete simulation on a single run without computers crashing for some reason I have been trying to divide my production runs in several stages that I therefore need to link.

To link the partial simulations (say a 100 time-steps each with a time-step of 5 fento seconds) I use the variable IRDWFK 1 to use the wavefunction from the previous partial run and update the positions and velocities in the new input file. ABINIT tells me that the file is good for restarting the simulation and takes it from there.

The "problem" is that the first time (MD step 1 not 0) ABINIT updates the positions and velocities using the Verlet algorithm it only gives me the first 6 digits of positions and velocities equal to those I would obtain if I had continued the simulation on a single run. Then as more and more integrations are performed the positions and velocities start to get more and more different from those I would obtain in a single run.

Is there any way of restarting a MD calculation in a such a way that the results are identical whether we do it in several stages using the _WFK files as input or in just a single run?

I would be highly appreciated if anyone could give some hint on this type of calculation as I am new to ABINIT and also to Hellman-Feynamnn MD.

Thanks in advance.

Nuno Galamba

-- 
Pierre-Matthieu Anglade